jiayi

8月23号之后全是重复的

推理Can't find 'adapter_config.json'

5

首先推理代码 ``` from peft import PeftModel model = PeftModel.from_pretrained(model, "./output/") ``` 不应该指定具体checkpoint文件吗？ 指定了之后，报错`Can't find 'adapter_config.json' at 'output/checkpoint-400'` 请问如何解决？

[ntbb-ladder.lib.php](https://github.com/smogon/pokemon-showdown-client/blob/8737e397ae4ca3734a0ec89dcf03692063057598/lib/ntbb-ladder.lib.php) line 423 shows the K factor is 40 if elo > 1300. But [ladderhelp.md](https://github.com/smogon/pokemon-showdown-client/blob/8737e397ae4ca3734a0ec89dcf03692063057598/pokemonshowdown.com/pages/ladderhelp.md) shows K = 32 if Elo is 1600 or higher The change was in [this...

Out of Memory running Qwen3-30B-A3B with 32k sequence length on 4 nodes (32 GPUs)

6

### Description I’m trying to train GRPO on the **Qwen3-30B-A3B** model with a sequence length of **32,000** tokens on **4 nodes** (total 32 GPUs with with 80 GB of VRAM),...